11th March 2026: Dr Carla Perez-Almendros

Spring semester, week 7

1.10-2pm [with optional extended discussion time until 2.20pm]

NLP for Subtle Bias Detection: From Patronizing Language to Mansplaining

Dr Carla Perez-Almendros (Cardiff University)

Language is never neutral. The words we choose, the tone we adopt, and the narratives we construct reflect and reinforce social hierarchies, power imbalances, and inequalities. This talk explores how Natural Language Processing (NLP) can help us identify and understand subtle forms of linguistic bias, particularly patronizing and condescending language (PCL) and mansplaining, that often go unrecognized but carry significant social consequences.

Drawing on my research into PCL detection and gender-based microaggressions, we will examine how certain linguistic patterns systematically diminish and marginalize vulnerable communities. Patronizing and condescending language operates through distinct techniques that convey superiority and pity, feeding stereotypes and reinforcing discrimination. Similarly, mansplaining represents a gendered variant of this phenomenon, where condescending discourse becomes a mechanism for asserting dominance over women’s knowledge and expertise.

This talk takes a linguistic and social perspective on these phenomena. We will try to answer questions such as: What linguistic features enable these harmful patterns? How do they function rhetorically to diminish others? And critically, can we use NLP insights to enhance our understanding of subtle bias in language?

I will share findings (and challenges!) from the creation of the Don’t Patronize Me! dataset and the “Well, actually…” corpus of mansplaining narratives. We will discuss what NLP-based language models (such as ChatGPT and others before) reveal and, crucially, what they fail to capture about these subtle forms of discrimination. We will see that the task of detecting bias in language is not merely a technical challenge; it is fundamentally about understanding how language embeds and perpetuates social inequalities, and how linguistic analysis can contribute to more equitable communication practices, even though we will be introducing our own biases along the process!

This session takes place in Room 3.58 of the John Percival Building [venue currently TBC] at Cardiff University or join us via Teams using this link.